X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fqede%2Fqede_rxtx.c;h=235f78ae358d1291ccb644a5089f70d430107c43;hb=35b2d13fd6fdcbd191f2a30d74648faeb1186c65;hp=f65c8338405a78553bf868c94ee4363fce1740b3;hpb=98a7ea332ba3da0f74ec951595d36a616165b255;p=dpdk.git diff --git a/drivers/net/qede/qede_rxtx.c b/drivers/net/qede/qede_rxtx.c index f65c833840..235f78ae35 100644 --- a/drivers/net/qede/qede_rxtx.c +++ b/drivers/net/qede/qede_rxtx.c @@ -1,9 +1,7 @@ -/* - * Copyright (c) 2016 QLogic Corporation. +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright (c) 2016 - 2018 Cavium Inc. * All rights reserved. - * www.qlogic.com - * - * See LICENSE.qede_pmd for copyright and licensing details. + * www.cavium.com */ #include @@ -28,7 +26,7 @@ static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq) } rxq->sw_rx_ring[idx].mbuf = new_mb; rxq->sw_rx_ring[idx].page_offset = 0; - mapping = rte_mbuf_data_dma_addr_default(new_mb); + mapping = rte_mbuf_data_iova_default(new_mb); /* Advance PROD and get BD pointer */ rx_bd = (struct eth_rx_bd *)ecore_chain_produce(&rxq->rx_bd_ring); rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping)); @@ -37,6 +35,95 @@ static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq) return 0; } +#define QEDE_MAX_BULK_ALLOC_COUNT 512 + +static inline int qede_alloc_rx_bulk_mbufs(struct qede_rx_queue *rxq, int count) +{ + void *obj_p[QEDE_MAX_BULK_ALLOC_COUNT] __rte_cache_aligned; + struct rte_mbuf *mbuf = NULL; + struct eth_rx_bd *rx_bd; + dma_addr_t mapping; + int i, ret = 0; + uint16_t idx; + + idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq); + + if (count > QEDE_MAX_BULK_ALLOC_COUNT) + count = QEDE_MAX_BULK_ALLOC_COUNT; + + ret = rte_mempool_get_bulk(rxq->mb_pool, obj_p, count); + if (unlikely(ret)) { + PMD_RX_LOG(ERR, rxq, + "Failed to allocate %d rx buffers " + "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u", + count, idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq), + rte_mempool_avail_count(rxq->mb_pool), + rte_mempool_in_use_count(rxq->mb_pool)); + return -ENOMEM; + } + + for (i = 0; i < count; i++) { + mbuf = obj_p[i]; + if (likely(i < count - 1)) + rte_prefetch0(obj_p[i + 1]); + + idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq); + rxq->sw_rx_ring[idx].mbuf = mbuf; + rxq->sw_rx_ring[idx].page_offset = 0; + mapping = rte_mbuf_data_iova_default(mbuf); + rx_bd = (struct eth_rx_bd *) + ecore_chain_produce(&rxq->rx_bd_ring); + rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping)); + rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping)); + rxq->sw_rx_prod++; + } + + return 0; +} + +/* Criterias for calculating Rx buffer size - + * 1) rx_buf_size should not exceed the size of mbuf + * 2) In scattered_rx mode - minimum rx_buf_size should be + * (MTU + Maximum L2 Header Size + 2) / ETH_RX_MAX_BUFF_PER_PKT + * 3) In regular mode - minimum rx_buf_size should be + * (MTU + Maximum L2 Header Size + 2) + * In above cases +2 corrosponds to 2 bytes padding in front of L2 + * header. + * 4) rx_buf_size should be cacheline-size aligned. So considering + * criteria 1, we need to adjust the size to floor instead of ceil, + * so that we don't exceed mbuf size while ceiling rx_buf_size. + */ +int +qede_calc_rx_buf_size(struct rte_eth_dev *dev, uint16_t mbufsz, + uint16_t max_frame_size) +{ + struct qede_dev *qdev = QEDE_INIT_QDEV(dev); + struct ecore_dev *edev = QEDE_INIT_EDEV(qdev); + int rx_buf_size; + + if (dev->data->scattered_rx) { + /* per HW limitation, only ETH_RX_MAX_BUFF_PER_PKT number of + * bufferes can be used for single packet. So need to make sure + * mbuf size is sufficient enough for this. + */ + if ((mbufsz * ETH_RX_MAX_BUFF_PER_PKT) < + (max_frame_size + QEDE_ETH_OVERHEAD)) { + DP_ERR(edev, "mbuf %d size is not enough to hold max fragments (%d) for max rx packet length (%d)\n", + mbufsz, ETH_RX_MAX_BUFF_PER_PKT, max_frame_size); + return -EINVAL; + } + + rx_buf_size = RTE_MAX(mbufsz, + (max_frame_size + QEDE_ETH_OVERHEAD) / + ETH_RX_MAX_BUFF_PER_PKT); + } else { + rx_buf_size = max_frame_size + QEDE_ETH_OVERHEAD; + } + + /* Align to cache-line size if needed */ + return QEDE_FLOOR_TO_CACHE_LINE_SIZE(rx_buf_size); +} + int qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, uint16_t nb_desc, unsigned int socket_id, @@ -82,24 +169,28 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, rxq->nb_rx_desc = nb_desc; rxq->queue_id = queue_idx; rxq->port_id = dev->data->port_id; + max_rx_pkt_len = (uint16_t)rxmode->max_rx_pkt_len; - qdev->mtu = max_rx_pkt_len; /* Fix up RX buffer size */ bufsz = (uint16_t)rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM; - if ((rxmode->enable_scatter) || + /* cache align the mbuf size to simplfy rx_buf_size calculation */ + bufsz = QEDE_FLOOR_TO_CACHE_LINE_SIZE(bufsz); + if ((rxmode->offloads & DEV_RX_OFFLOAD_SCATTER) || (max_rx_pkt_len + QEDE_ETH_OVERHEAD) > bufsz) { if (!dev->data->scattered_rx) { DP_INFO(edev, "Forcing scatter-gather mode\n"); dev->data->scattered_rx = 1; } } - if (dev->data->scattered_rx) - rxq->rx_buf_size = bufsz + QEDE_ETH_OVERHEAD; - else - rxq->rx_buf_size = qdev->mtu + QEDE_ETH_OVERHEAD; - /* Align to cache-line size if needed */ - rxq->rx_buf_size = QEDE_CEIL_TO_CACHE_LINE_SIZE(rxq->rx_buf_size); + + rc = qede_calc_rx_buf_size(dev, bufsz, max_rx_pkt_len); + if (rc < 0) { + rte_free(rxq); + return rc; + } + + rxq->rx_buf_size = rc; DP_INFO(edev, "mtu %u mbufsz %u bd_max_bytes %u scatter_mode %d\n", qdev->mtu, bufsz, rxq->rx_buf_size, dev->data->scattered_rx); @@ -109,9 +200,8 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, rxq->sw_rx_ring = rte_zmalloc_socket("sw_rx_ring", size, RTE_CACHE_LINE_SIZE, socket_id); if (!rxq->sw_rx_ring) { - DP_NOTICE(edev, false, - "Unable to alloc memory for sw_rx_ring on socket %u\n", - socket_id); + DP_ERR(edev, "Memory allocation fails for sw_rx_ring on" + " socket %u\n", socket_id); rte_free(rxq); return -ENOMEM; } @@ -127,9 +217,8 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, NULL); if (rc != ECORE_SUCCESS) { - DP_NOTICE(edev, false, - "Unable to alloc memory for rxbd ring on socket %u\n", - socket_id); + DP_ERR(edev, "Memory allocation fails for RX BD ring" + " on socket %u\n", socket_id); rte_free(rxq->sw_rx_ring); rte_free(rxq); return -ENOMEM; @@ -146,10 +235,9 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, NULL); if (rc != ECORE_SUCCESS) { - DP_NOTICE(edev, false, - "Unable to alloc memory for cqe ring on socket %u\n", - socket_id); - /* TBD: Freeing RX BD ring */ + DP_ERR(edev, "Memory allocation fails for RX CQE ring" + " on socket %u\n", socket_id); + qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring); rte_free(rxq->sw_rx_ring); rte_free(rxq); return -ENOMEM; @@ -159,7 +247,7 @@ qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx, qdev->fp_array[queue_idx].rxq = rxq; DP_INFO(edev, "rxq %d num_desc %u rx_buf_size=%u socket %u\n", - queue_idx, nb_desc, qdev->mtu, socket_id); + queue_idx, nb_desc, rxq->rx_buf_size, socket_id); return 0; } @@ -193,9 +281,16 @@ static void qede_rx_queue_release_mbufs(struct qede_rx_queue *rxq) void qede_rx_queue_release(void *rx_queue) { struct qede_rx_queue *rxq = rx_queue; + struct qede_dev *qdev; + struct ecore_dev *edev; if (rxq) { + qdev = rxq->qdev; + edev = QEDE_INIT_EDEV(qdev); + PMD_INIT_FUNC_TRACE(edev); qede_rx_queue_release_mbufs(rxq); + qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring); + qdev->ops->common->chain_free(edev, &rxq->rx_comp_ring); rte_free(rxq->sw_rx_ring); rte_free(rxq); } @@ -300,6 +395,7 @@ qede_tx_queue_setup(struct rte_eth_dev *dev, DP_ERR(edev, "Unable to allocate memory for txbd ring on socket %u", socket_id); + qdev->ops->common->chain_free(edev, &txq->tx_pbl); qede_tx_queue_release(txq); return -ENOMEM; } @@ -350,9 +446,15 @@ static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq) void qede_tx_queue_release(void *tx_queue) { struct qede_tx_queue *txq = tx_queue; + struct qede_dev *qdev; + struct ecore_dev *edev; if (txq) { + qdev = txq->qdev; + edev = QEDE_INIT_EDEV(qdev); + PMD_INIT_FUNC_TRACE(edev); qede_tx_queue_release_mbufs(txq); + qdev->ops->common->chain_free(edev, &txq->tx_pbl); rte_free(txq->sw_tx_ring); rte_free(txq); } @@ -363,23 +465,23 @@ static int qede_alloc_mem_sb(struct qede_dev *qdev, struct ecore_sb_info *sb_info, uint16_t sb_id) { - struct ecore_dev *edev = &qdev->edev; - struct status_block *sb_virt; + struct ecore_dev *edev = QEDE_INIT_EDEV(qdev); + struct status_block_e4 *sb_virt; dma_addr_t sb_phys; int rc; - sb_virt = OSAL_DMA_ALLOC_COHERENT(edev, &sb_phys, sizeof(*sb_virt)); - + sb_virt = OSAL_DMA_ALLOC_COHERENT(edev, &sb_phys, + sizeof(struct status_block_e4)); if (!sb_virt) { DP_ERR(edev, "Status block allocation failed\n"); return -ENOMEM; } - rc = qdev->ops->common->sb_init(edev, sb_info, sb_virt, sb_phys, sb_id); if (rc) { DP_ERR(edev, "Status block initialization failed\n"); - /* TBD: No dma_free_coherent possible */ + OSAL_DMA_FREE_COHERENT(edev, sb_virt, sb_phys, + sizeof(struct status_block_e4)); return rc; } @@ -417,6 +519,8 @@ int qede_alloc_fp_resc(struct qede_dev *qdev) for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) { fp = &qdev->fp_array[sb_idx]; + if (!fp) + continue; fp->sb_info = rte_calloc("sb", 1, sizeof(struct ecore_sb_info), RTE_CACHE_LINE_SIZE); if (!fp->sb_info) { @@ -437,20 +541,43 @@ int qede_alloc_fp_resc(struct qede_dev *qdev) void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev) { struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev); - __rte_unused struct ecore_dev *edev = QEDE_INIT_EDEV(qdev); + struct ecore_dev *edev = QEDE_INIT_EDEV(qdev); struct qede_fastpath *fp; uint16_t sb_idx; + uint8_t i; PMD_INIT_FUNC_TRACE(edev); for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) { fp = &qdev->fp_array[sb_idx]; + if (!fp) + continue; DP_INFO(edev, "Free sb_info index 0x%x\n", fp->sb_info->igu_sb_id); - if (fp->sb_info) + if (fp->sb_info) { + OSAL_DMA_FREE_COHERENT(edev, fp->sb_info->sb_virt, + fp->sb_info->sb_phys, + sizeof(struct status_block_e4)); rte_free(fp->sb_info); - fp->sb_info = NULL; + fp->sb_info = NULL; + } + } + + /* Free packet buffers and ring memories */ + for (i = 0; i < eth_dev->data->nb_rx_queues; i++) { + if (eth_dev->data->rx_queues[i]) { + qede_rx_queue_release(eth_dev->data->rx_queues[i]); + eth_dev->data->rx_queues[i] = NULL; + } } + + for (i = 0; i < eth_dev->data->nb_tx_queues; i++) { + if (eth_dev->data->tx_queues[i]) { + qede_tx_queue_release(eth_dev->data->tx_queues[i]); + eth_dev->data->tx_queues[i] = NULL; + } + } + if (qdev->fp_array) rte_free(qdev->fp_array); qdev->fp_array = NULL; @@ -521,9 +648,10 @@ qede_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id) ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0); /* Prepare ramrod */ memset(¶ms, 0, sizeof(params)); - params.queue_id = rx_queue_id; + params.queue_id = rx_queue_id / edev->num_hwfns; params.vport_id = 0; - params.sb = fp->sb_info->igu_sb_id; + params.stats_id = params.vport_id; + params.p_sb = fp->sb_info; DP_INFO(edev, "rxq %u igu_sb_id 0x%x\n", fp->rxq->queue_id, fp->sb_info->igu_sb_id); params.sb_idx = RX_PI; @@ -579,9 +707,10 @@ qede_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id) txq = eth_dev->data->tx_queues[tx_queue_id]; fp = &qdev->fp_array[tx_queue_id]; memset(¶ms, 0, sizeof(params)); - params.queue_id = tx_queue_id; + params.queue_id = tx_queue_id / edev->num_hwfns; params.vport_id = 0; - params.sb = fp->sb_info->igu_sb_id; + params.stats_id = params.vport_id; + params.p_sb = fp->sb_info; DP_INFO(edev, "txq %u igu_sb_id 0x%x\n", fp->txq->queue_id, fp->sb_info->igu_sb_id); params.sb_idx = TX_PI(0); /* tc = 0 */ @@ -672,7 +801,6 @@ qede_process_tx_compl(__rte_unused struct ecore_dev *edev, qede_free_tx_pkt(txq); } - static int qede_drain_txq(struct qede_dev *qdev, struct qede_tx_queue *txq, bool allow_drain) { @@ -708,7 +836,6 @@ static int qede_drain_txq(struct qede_dev *qdev, return 0; } - /* Stops a given TX queue in the HW */ static int qede_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id) { @@ -749,7 +876,7 @@ int qede_start_queues(struct rte_eth_dev *eth_dev) { struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev); uint8_t id; - int rc; + int rc = -1; for_each_rss(id) { rc = qede_rx_queue_start(eth_dev, id); @@ -781,12 +908,18 @@ void qede_stop_queues(struct rte_eth_dev *eth_dev) } } -static bool qede_tunn_exist(uint16_t flag) +static inline bool qede_tunn_exist(uint16_t flag) { return !!((PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK << PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT) & flag); } +static inline uint8_t qede_check_tunn_csum_l3(uint16_t flag) +{ + return !!((PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_MASK << + PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_SHIFT) & flag); +} + /* * qede_check_tunn_csum_l4: * Returns: @@ -813,6 +946,127 @@ static inline uint8_t qede_check_notunn_csum_l4(uint16_t flag) return 0; } +/* Returns outer L2, L3 and L4 packet_type for tunneled packets */ +static inline uint32_t qede_rx_cqe_to_pkt_type_outer(struct rte_mbuf *m) +{ + uint32_t packet_type = RTE_PTYPE_UNKNOWN; + struct rte_ether_hdr *eth_hdr; + struct ipv4_hdr *ipv4_hdr; + struct ipv6_hdr *ipv6_hdr; + struct rte_vlan_hdr *vlan_hdr; + uint16_t ethertype; + bool vlan_tagged = 0; + uint16_t len; + + eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); + len = sizeof(struct rte_ether_hdr); + ethertype = rte_cpu_to_be_16(eth_hdr->ether_type); + + /* Note: Valid only if VLAN stripping is disabled */ + if (ethertype == RTE_ETHER_TYPE_VLAN) { + vlan_tagged = 1; + vlan_hdr = (struct rte_vlan_hdr *)(eth_hdr + 1); + len += sizeof(struct rte_vlan_hdr); + ethertype = rte_cpu_to_be_16(vlan_hdr->eth_proto); + } + + if (ethertype == RTE_ETHER_TYPE_IPv4) { + packet_type |= RTE_PTYPE_L3_IPV4; + ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *, len); + if (ipv4_hdr->next_proto_id == IPPROTO_TCP) + packet_type |= RTE_PTYPE_L4_TCP; + else if (ipv4_hdr->next_proto_id == IPPROTO_UDP) + packet_type |= RTE_PTYPE_L4_UDP; + } else if (ethertype == RTE_ETHER_TYPE_IPv6) { + packet_type |= RTE_PTYPE_L3_IPV6; + ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *, len); + if (ipv6_hdr->proto == IPPROTO_TCP) + packet_type |= RTE_PTYPE_L4_TCP; + else if (ipv6_hdr->proto == IPPROTO_UDP) + packet_type |= RTE_PTYPE_L4_UDP; + } + + if (vlan_tagged) + packet_type |= RTE_PTYPE_L2_ETHER_VLAN; + else + packet_type |= RTE_PTYPE_L2_ETHER; + + return packet_type; +} + +static inline uint32_t qede_rx_cqe_to_pkt_type_inner(uint16_t flags) +{ + uint16_t val; + + /* Lookup table */ + static const uint32_t + ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = { + [QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_INNER_L3_IPV4 | + RTE_PTYPE_INNER_L2_ETHER, + [QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_INNER_L3_IPV6 | + RTE_PTYPE_INNER_L2_ETHER, + [QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_INNER_L3_IPV4 | + RTE_PTYPE_INNER_L4_TCP | + RTE_PTYPE_INNER_L2_ETHER, + [QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_INNER_L3_IPV6 | + RTE_PTYPE_INNER_L4_TCP | + RTE_PTYPE_INNER_L2_ETHER, + [QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_INNER_L3_IPV4 | + RTE_PTYPE_INNER_L4_UDP | + RTE_PTYPE_INNER_L2_ETHER, + [QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_INNER_L3_IPV6 | + RTE_PTYPE_INNER_L4_UDP | + RTE_PTYPE_INNER_L2_ETHER, + /* Frags with no VLAN */ + [QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_INNER_L3_IPV4 | + RTE_PTYPE_INNER_L4_FRAG | + RTE_PTYPE_INNER_L2_ETHER, + [QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_INNER_L3_IPV6 | + RTE_PTYPE_INNER_L4_FRAG | + RTE_PTYPE_INNER_L2_ETHER, + /* VLANs */ + [QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_INNER_L3_IPV4 | + RTE_PTYPE_INNER_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_INNER_L3_IPV6 | + RTE_PTYPE_INNER_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV4 | + RTE_PTYPE_INNER_L4_TCP | + RTE_PTYPE_INNER_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV6 | + RTE_PTYPE_INNER_L4_TCP | + RTE_PTYPE_INNER_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV4 | + RTE_PTYPE_INNER_L4_UDP | + RTE_PTYPE_INNER_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV6 | + RTE_PTYPE_INNER_L4_UDP | + RTE_PTYPE_INNER_L2_ETHER_VLAN, + /* Frags with VLAN */ + [QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV4 | + RTE_PTYPE_INNER_L4_FRAG | + RTE_PTYPE_INNER_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV6 | + RTE_PTYPE_INNER_L4_FRAG | + RTE_PTYPE_INNER_L2_ETHER_VLAN, + }; + + /* Bits (0..3) provides L3/L4 protocol type */ + /* Bits (4,5) provides frag and VLAN info */ + val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK << + PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) | + (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK << + PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) | + (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK << + PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) | + (PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK << + PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags; + + if (val < QEDE_PKT_TYPE_MAX) + return ptype_lkup_tbl[val]; + + return RTE_PTYPE_UNKNOWN; +} + static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags) { uint16_t val; @@ -820,24 +1074,68 @@ static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags) /* Lookup table */ static const uint32_t ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = { - [QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4, - [QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6, - [QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP, - [QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP, - [QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP, - [QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP, + [QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4 | + RTE_PTYPE_L4_TCP | + RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6 | + RTE_PTYPE_L4_TCP | + RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4 | + RTE_PTYPE_L4_UDP | + RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6 | + RTE_PTYPE_L4_UDP | + RTE_PTYPE_L2_ETHER, + /* Frags with no VLAN */ + [QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_L3_IPV4 | + RTE_PTYPE_L4_FRAG | + RTE_PTYPE_L2_ETHER, + [QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_L3_IPV6 | + RTE_PTYPE_L4_FRAG | + RTE_PTYPE_L2_ETHER, + /* VLANs */ + [QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_L3_IPV4 | + RTE_PTYPE_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_L3_IPV6 | + RTE_PTYPE_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_L3_IPV4 | + RTE_PTYPE_L4_TCP | + RTE_PTYPE_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_L3_IPV6 | + RTE_PTYPE_L4_TCP | + RTE_PTYPE_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_L3_IPV4 | + RTE_PTYPE_L4_UDP | + RTE_PTYPE_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_L3_IPV6 | + RTE_PTYPE_L4_UDP | + RTE_PTYPE_L2_ETHER_VLAN, + /* Frags with VLAN */ + [QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_L3_IPV4 | + RTE_PTYPE_L4_FRAG | + RTE_PTYPE_L2_ETHER_VLAN, + [QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_L3_IPV6 | + RTE_PTYPE_L4_FRAG | + RTE_PTYPE_L2_ETHER_VLAN, }; /* Bits (0..3) provides L3/L4 protocol type */ + /* Bits (4,5) provides frag and VLAN info */ val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK << PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) | (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK << - PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT)) & flags; + PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) | + (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK << + PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) | + (PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK << + PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags; if (val < QEDE_PKT_TYPE_MAX) - return ptype_lkup_tbl[val] | RTE_PTYPE_L2_ETHER; - else - return RTE_PTYPE_UNKNOWN; + return ptype_lkup_tbl[val]; + + return RTE_PTYPE_UNKNOWN; } static inline uint8_t @@ -855,7 +1153,7 @@ qede_check_notunn_csum_l3(struct rte_mbuf *m, uint16_t flag) m->packet_type = qede_rx_cqe_to_pkt_type(flag); if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) { ip = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *, - sizeof(struct ether_hdr)); + sizeof(struct rte_ether_hdr)); pkt_csum = ip->hdr_checksum; ip->hdr_checksum = 0; calc_csum = rte_ipv4_cksum(ip); @@ -879,14 +1177,14 @@ qede_reuse_page(__rte_unused struct qede_dev *qdev, struct qede_rx_queue *rxq, struct qede_rx_entry *curr_cons) { struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring); - uint16_t idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq); + uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq); struct qede_rx_entry *curr_prod; dma_addr_t new_mapping; curr_prod = &rxq->sw_rx_ring[idx]; *curr_prod = *curr_cons; - new_mapping = rte_mbuf_data_dma_addr_default(curr_prod->mbuf) + + new_mapping = rte_mbuf_data_iova_default(curr_prod->mbuf) + curr_prod->page_offset; rx_bd_prod->addr.hi = rte_cpu_to_le_32(U64_HI(new_mapping)); @@ -985,17 +1283,17 @@ static inline uint32_t qede_rx_cqe_to_tunn_pkt_type(uint16_t flags) [QEDE_PKT_TYPE_TUNN_GRE] = RTE_PTYPE_TUNNEL_GRE, [QEDE_PKT_TYPE_TUNN_VXLAN] = RTE_PTYPE_TUNNEL_VXLAN, [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GENEVE] = - RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L2_ETHER, + RTE_PTYPE_TUNNEL_GENEVE, [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GRE] = - RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L2_ETHER, + RTE_PTYPE_TUNNEL_GRE, [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_VXLAN] = - RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L2_ETHER, + RTE_PTYPE_TUNNEL_VXLAN, [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GENEVE] = - RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L2_ETHER, + RTE_PTYPE_TUNNEL_GENEVE, [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GRE] = - RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L2_ETHER, + RTE_PTYPE_TUNNEL_GRE, [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_VXLAN] = - RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L2_ETHER, + RTE_PTYPE_TUNNEL_VXLAN, [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GENEVE] = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4, [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GRE] = @@ -1051,7 +1349,7 @@ qede_process_sg_pkts(void *p_rxq, struct rte_mbuf *rx_mb, pkt_len; if (unlikely(!cur_size)) { PMD_RX_LOG(ERR, rxq, "Length is 0 while %u BDs" - " left for mapping jumbo", num_segs); + " left for mapping jumbo\n", num_segs); qede_recycle_rx_bd_ring(rxq, qdev, num_segs); return -EINVAL; } @@ -1069,6 +1367,27 @@ qede_process_sg_pkts(void *p_rxq, struct rte_mbuf *rx_mb, return 0; } +#ifdef RTE_LIBRTE_QEDE_DEBUG_RX +static inline void +print_rx_bd_info(struct rte_mbuf *m, struct qede_rx_queue *rxq, + uint8_t bitfield) +{ + PMD_RX_LOG(INFO, rxq, + "len 0x%04x bf 0x%04x hash_val 0x%x" + " ol_flags 0x%04lx l2=%s l3=%s l4=%s tunn=%s" + " inner_l2=%s inner_l3=%s inner_l4=%s\n", + m->data_len, bitfield, m->hash.rss, + (unsigned long)m->ol_flags, + rte_get_ptype_l2_name(m->packet_type), + rte_get_ptype_l3_name(m->packet_type), + rte_get_ptype_l4_name(m->packet_type), + rte_get_ptype_tunnel_name(m->packet_type), + rte_get_ptype_inner_l2_name(m->packet_type), + rte_get_ptype_inner_l3_name(m->packet_type), + rte_get_ptype_inner_l4_name(m->packet_type)); +} +#endif + uint16_t qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) { @@ -1089,10 +1408,8 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) uint16_t parse_flag; #ifdef RTE_LIBRTE_QEDE_DEBUG_RX uint8_t bitfield_val; - enum rss_hash_type htype; #endif uint8_t tunn_parse_flag; - uint8_t j; struct eth_fast_path_rx_tpa_start_cqe *cqe_start_tpa; uint64_t ol_flags; uint32_t packet_type; @@ -1101,6 +1418,27 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) uint8_t offset, tpa_agg_idx, flags; struct qede_agg_info *tpa_info = NULL; uint32_t rss_hash; + int rx_alloc_count = 0; + + + /* Allocate buffers that we used in previous loop */ + if (rxq->rx_alloc_count) { + if (unlikely(qede_alloc_rx_bulk_mbufs(rxq, + rxq->rx_alloc_count))) { + struct rte_eth_dev *dev; + + PMD_RX_LOG(ERR, rxq, + "New buffer allocation failed," + "dropping incoming packetn"); + dev = &rte_eth_devices[rxq->port_id]; + dev->data->rx_mbuf_alloc_failed += + rxq->rx_alloc_count; + rxq->rx_alloc_errors += rxq->rx_alloc_count; + return 0; + } + qede_update_rx_prod(qdev, rxq); + rxq->rx_alloc_count = 0; + } hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr); sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring); @@ -1183,8 +1521,6 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash); #ifdef RTE_LIBRTE_QEDE_DEBUG_RX bitfield_val = fp_cqe->bitfields; - htype = (uint8_t)GET_FIELD(bitfield_val, - ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE); #endif } else { parse_flag = @@ -1195,8 +1531,6 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) vlan_tci = rte_le_to_cpu_16(cqe_start_tpa->vlan_tag); #ifdef RTE_LIBRTE_QEDE_DEBUG_RX bitfield_val = cqe_start_tpa->bitfields; - htype = (uint8_t)GET_FIELD(bitfield_val, - ETH_FAST_PATH_RX_TPA_START_CQE_RSS_HASH_TYPE); #endif rss_hash = rte_le_to_cpu_32(cqe_start_tpa->rss_hash); } @@ -1210,71 +1544,82 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) ol_flags |= PKT_RX_L4_CKSUM_BAD; } else { ol_flags |= PKT_RX_L4_CKSUM_GOOD; - if (tpa_start_flg) - flags = - cqe_start_tpa->tunnel_pars_flags.flags; - else - flags = fp_cqe->tunnel_pars_flags.flags; - tunn_parse_flag = flags; - packet_type = - qede_rx_cqe_to_tunn_pkt_type(tunn_parse_flag); } - } else { - PMD_RX_LOG(INFO, rxq, "Rx non-tunneled packet\n"); - if (unlikely(qede_check_notunn_csum_l4(parse_flag))) { - PMD_RX_LOG(ERR, rxq, - "L4 csum failed, flags = 0x%x\n", - parse_flag); - rxq->rx_hw_errors++; - ol_flags |= PKT_RX_L4_CKSUM_BAD; - } else { - ol_flags |= PKT_RX_L4_CKSUM_GOOD; - } - if (unlikely(qede_check_notunn_csum_l3(rx_mb, - parse_flag))) { + + if (unlikely(qede_check_tunn_csum_l3(parse_flag))) { PMD_RX_LOG(ERR, rxq, - "IP csum failed, flags = 0x%x\n", - parse_flag); - rxq->rx_hw_errors++; - ol_flags |= PKT_RX_IP_CKSUM_BAD; + "Outer L3 csum failed, flags = 0x%x\n", + parse_flag); + rxq->rx_hw_errors++; + ol_flags |= PKT_RX_EIP_CKSUM_BAD; } else { - ol_flags |= PKT_RX_IP_CKSUM_GOOD; - packet_type = - qede_rx_cqe_to_pkt_type(parse_flag); + ol_flags |= PKT_RX_IP_CKSUM_GOOD; } + + if (tpa_start_flg) + flags = cqe_start_tpa->tunnel_pars_flags.flags; + else + flags = fp_cqe->tunnel_pars_flags.flags; + tunn_parse_flag = flags; + + /* Tunnel_type */ + packet_type = + qede_rx_cqe_to_tunn_pkt_type(tunn_parse_flag); + + /* Inner header */ + packet_type |= + qede_rx_cqe_to_pkt_type_inner(parse_flag); + + /* Outer L3/L4 types is not available in CQE */ + packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb); + + /* Outer L3/L4 types is not available in CQE. + * Need to add offset to parse correctly, + */ + rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM; + packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb); + } else { + packet_type |= qede_rx_cqe_to_pkt_type(parse_flag); } - if (CQE_HAS_VLAN(parse_flag)) { - ol_flags |= PKT_RX_VLAN_PKT; - if (qdev->vlan_strip_flg) { - ol_flags |= PKT_RX_VLAN_STRIPPED; - rx_mb->vlan_tci = vlan_tci; - } + /* Common handling for non-tunnel packets and for inner + * headers in the case of tunnel. + */ + if (unlikely(qede_check_notunn_csum_l4(parse_flag))) { + PMD_RX_LOG(ERR, rxq, + "L4 csum failed, flags = 0x%x\n", + parse_flag); + rxq->rx_hw_errors++; + ol_flags |= PKT_RX_L4_CKSUM_BAD; + } else { + ol_flags |= PKT_RX_L4_CKSUM_GOOD; } - if (CQE_HAS_OUTER_VLAN(parse_flag)) { - ol_flags |= PKT_RX_QINQ_PKT; + if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) { + PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n", + parse_flag); + rxq->rx_hw_errors++; + ol_flags |= PKT_RX_IP_CKSUM_BAD; + } else { + ol_flags |= PKT_RX_IP_CKSUM_GOOD; + } + + if (CQE_HAS_VLAN(parse_flag) || + CQE_HAS_OUTER_VLAN(parse_flag)) { + /* Note: FW doesn't indicate Q-in-Q packet */ + ol_flags |= PKT_RX_VLAN; if (qdev->vlan_strip_flg) { + ol_flags |= PKT_RX_VLAN_STRIPPED; rx_mb->vlan_tci = vlan_tci; - ol_flags |= PKT_RX_QINQ_STRIPPED; } - rx_mb->vlan_tci_outer = 0; } + /* RSS Hash */ if (qdev->rss_enable) { ol_flags |= PKT_RX_RSS_HASH; rx_mb->hash.rss = rss_hash; } - if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) { - PMD_RX_LOG(ERR, rxq, - "New buffer allocation failed," - "dropping incoming packet\n"); - qede_recycle_rx_bd_ring(rxq, qdev, fp_cqe->bd_num); - rte_eth_devices[rxq->port_id]. - data->rx_mbuf_alloc_failed++; - rxq->rx_alloc_errors++; - break; - } + rx_alloc_count++; qede_rx_bd_ring_consume(rxq); if (!tpa_start_flg && fp_cqe->bd_num > 1) { @@ -1286,17 +1631,9 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) if (qede_process_sg_pkts(p_rxq, seg1, num_segs, pkt_len - len)) goto next_cqe; - for (j = 0; j < num_segs; j++) { - if (qede_alloc_rx_buffer(rxq)) { - PMD_RX_LOG(ERR, rxq, - "Buffer allocation failed"); - rte_eth_devices[rxq->port_id]. - data->rx_mbuf_alloc_failed++; - rxq->rx_alloc_errors++; - break; - } - rxq->rx_segs++; - } + + rx_alloc_count += num_segs; + rxq->rx_segs += num_segs; } rxq->rx_segs++; /* for the first segment */ @@ -1310,11 +1647,9 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) rx_mb->ol_flags = ol_flags; rx_mb->data_len = len; rx_mb->packet_type = packet_type; - PMD_RX_LOG(INFO, rxq, - "pkt_type 0x%04x len %u hash_type %d hash_val 0x%x" - " ol_flags 0x%04lx\n", - packet_type, len, htype, rx_mb->hash.rss, - (unsigned long)ol_flags); +#ifdef RTE_LIBRTE_QEDE_DEBUG_RX + print_rx_bd_info(rx_mb, rxq, bitfield_val); +#endif if (!tpa_start_flg) { rx_mb->nb_segs = fp_cqe->bd_num; rx_mb->pkt_len = pkt_len; @@ -1340,7 +1675,8 @@ next_cqe: } } - qede_update_rx_prod(qdev, rxq); + /* Request number of bufferes to be allocated in next loop */ + rxq->rx_alloc_count = rx_alloc_count; rxq->rcv_pkts += rx_pkt; @@ -1351,35 +1687,36 @@ next_cqe: /* Populate scatter gather buffer descriptor fields */ -static inline uint8_t +static inline uint16_t qede_encode_sg_bd(struct qede_tx_queue *p_txq, struct rte_mbuf *m_seg, - struct eth_tx_2nd_bd **bd2, struct eth_tx_3rd_bd **bd3) + struct eth_tx_2nd_bd **bd2, struct eth_tx_3rd_bd **bd3, + uint16_t start_seg) { struct qede_tx_queue *txq = p_txq; struct eth_tx_bd *tx_bd = NULL; dma_addr_t mapping; - uint8_t nb_segs = 0; + uint16_t nb_segs = 0; /* Check for scattered buffers */ while (m_seg) { - if (nb_segs == 0) { + if (start_seg == 0) { if (!*bd2) { *bd2 = (struct eth_tx_2nd_bd *) ecore_chain_produce(&txq->tx_pbl); memset(*bd2, 0, sizeof(struct eth_tx_2nd_bd)); nb_segs++; } - mapping = rte_mbuf_data_dma_addr(m_seg); + mapping = rte_mbuf_data_iova(m_seg); QEDE_BD_SET_ADDR_LEN(*bd2, mapping, m_seg->data_len); PMD_TX_LOG(DEBUG, txq, "BD2 len %04x", m_seg->data_len); - } else if (nb_segs == 1) { + } else if (start_seg == 1) { if (!*bd3) { *bd3 = (struct eth_tx_3rd_bd *) ecore_chain_produce(&txq->tx_pbl); memset(*bd3, 0, sizeof(struct eth_tx_3rd_bd)); nb_segs++; } - mapping = rte_mbuf_data_dma_addr(m_seg); + mapping = rte_mbuf_data_iova(m_seg); QEDE_BD_SET_ADDR_LEN(*bd3, mapping, m_seg->data_len); PMD_TX_LOG(DEBUG, txq, "BD3 len %04x", m_seg->data_len); } else { @@ -1387,10 +1724,11 @@ qede_encode_sg_bd(struct qede_tx_queue *p_txq, struct rte_mbuf *m_seg, ecore_chain_produce(&txq->tx_pbl); memset(tx_bd, 0, sizeof(*tx_bd)); nb_segs++; - mapping = rte_mbuf_data_dma_addr(m_seg); + mapping = rte_mbuf_data_iova(m_seg); QEDE_BD_SET_ADDR_LEN(tx_bd, mapping, m_seg->data_len); PMD_TX_LOG(DEBUG, txq, "BD len %04x", m_seg->data_len); } + start_seg++; m_seg = m_seg->next; } @@ -1410,20 +1748,24 @@ print_tx_bd_info(struct qede_tx_queue *txq, if (bd1) PMD_TX_LOG(INFO, txq, - "BD1: nbytes=%u nbds=%u bd_flags=%04x bf=%04x", - rte_cpu_to_le_16(bd1->nbytes), bd1->data.nbds, - bd1->data.bd_flags.bitfields, - rte_cpu_to_le_16(bd1->data.bitfields)); + "BD1: nbytes=0x%04x nbds=0x%04x bd_flags=0x%04x bf=0x%04x", + rte_cpu_to_le_16(bd1->nbytes), bd1->data.nbds, + bd1->data.bd_flags.bitfields, + rte_cpu_to_le_16(bd1->data.bitfields)); if (bd2) PMD_TX_LOG(INFO, txq, - "BD2: nbytes=%u bf=%04x\n", - rte_cpu_to_le_16(bd2->nbytes), bd2->data.bitfields1); + "BD2: nbytes=0x%04x bf1=0x%04x bf2=0x%04x tunn_ip=0x%04x\n", + rte_cpu_to_le_16(bd2->nbytes), bd2->data.bitfields1, + bd2->data.bitfields2, bd2->data.tunn_ip_size); if (bd3) PMD_TX_LOG(INFO, txq, - "BD3: nbytes=%u bf=%04x mss=%u\n", - rte_cpu_to_le_16(bd3->nbytes), - rte_cpu_to_le_16(bd3->data.bitfields), - rte_cpu_to_le_16(bd3->data.lso_mss)); + "BD3: nbytes=0x%04x bf=0x%04x MSS=0x%04x " + "tunn_l4_hdr_start_offset_w=0x%04x tunn_hdr_size=0x%04x\n", + rte_cpu_to_le_16(bd3->nbytes), + rte_cpu_to_le_16(bd3->data.bitfields), + rte_cpu_to_le_16(bd3->data.lso_mss), + bd3->data.tunn_l4_hdr_start_offset_w, + bd3->data.tunn_hdr_size_w); rte_get_tx_ol_flag_list(tx_ol_flags, ol_buf, sizeof(ol_buf)); PMD_TX_LOG(INFO, txq, "TX offloads = %s\n", ol_buf); @@ -1445,7 +1787,9 @@ qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts, uint64_t ol_flags; struct rte_mbuf *m; uint16_t i; +#ifdef RTE_LIBRTE_ETHDEV_DEBUG int ret; +#endif for (i = 0; i < nb_pkts; i++) { m = tx_pkts[i]; @@ -1467,6 +1811,18 @@ qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts, } } if (ol_flags & QEDE_TX_OFFLOAD_NOTSUP_MASK) { + /* We support only limited tunnel protocols */ + if (ol_flags & PKT_TX_TUNNEL_MASK) { + uint64_t temp; + + temp = ol_flags & PKT_TX_TUNNEL_MASK; + if (temp == PKT_TX_TUNNEL_VXLAN || + temp == PKT_TX_TUNNEL_GENEVE || + temp == PKT_TX_TUNNEL_MPLSINUDP || + temp == PKT_TX_TUNNEL_GRE) + continue; + } + rte_errno = -ENOTSUP; break; } @@ -1478,14 +1834,6 @@ qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts, break; } #endif - /* TBD: pseudo csum calcuation required if - * ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE not set? - */ - ret = rte_net_intel_cksum_prepare(m); - if (ret != 0) { - rte_errno = ret; - break; - } } #ifdef RTE_LIBRTE_QEDE_DEBUG_TX @@ -1496,6 +1844,27 @@ qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts, return i; } +#define MPLSINUDP_HDR_SIZE (12) + +#ifdef RTE_LIBRTE_QEDE_DEBUG_TX +static inline void +qede_mpls_tunn_tx_sanity_check(struct rte_mbuf *mbuf, + struct qede_tx_queue *txq) +{ + if (((mbuf->outer_l2_len + mbuf->outer_l3_len) / 2) > 0xff) + PMD_TX_LOG(ERR, txq, "tunn_l4_hdr_start_offset overflow\n"); + if (((mbuf->outer_l2_len + mbuf->outer_l3_len + + MPLSINUDP_HDR_SIZE) / 2) > 0xff) + PMD_TX_LOG(ERR, txq, "tunn_hdr_size overflow\n"); + if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE) / 2) > + ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK) + PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n"); + if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2) > + ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK) + PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n"); +} +#endif + uint16_t qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) { @@ -1510,9 +1879,10 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) uint16_t nb_frags; uint16_t nb_pkt_sent = 0; uint8_t nbds; - bool ipv6_ext_flg; bool lso_flg; + bool mplsoudp_flg; __rte_unused bool tunn_flg; + bool tunn_ipv6_ext_flg; struct eth_tx_1st_bd *bd1; struct eth_tx_2nd_bd *bd2; struct eth_tx_3rd_bd *bd3; @@ -1529,6 +1899,10 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) uint16_t mss; uint16_t bd3_bf; + uint8_t tunn_l4_hdr_start_offset; + uint8_t tunn_hdr_size; + uint8_t inner_l2_hdr_size; + uint16_t inner_l4_hdr_offset; if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) { PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u", @@ -1540,7 +1914,6 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl)); while (nb_tx_pkts--) { /* Init flags/values */ - ipv6_ext_flg = false; tunn_flg = false; lso_flg = false; nbds = 0; @@ -1555,6 +1928,10 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) bd2_bf2 = 0; mss = 0; bd3_bf = 0; + mplsoudp_flg = false; + tunn_ipv6_ext_flg = false; + tunn_hdr_size = 0; + tunn_l4_hdr_start_offset = 0; mbuf = *tx_pkts++; assert(mbuf); @@ -1566,21 +1943,18 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) tx_ol_flags = mbuf->ol_flags; bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT; -#define RTE_ETH_IS_IPV6_HDR_EXT(ptype) ((ptype) & RTE_PTYPE_L3_IPV6_EXT) - if (RTE_ETH_IS_IPV6_HDR_EXT(mbuf->packet_type)) { - ipv6_ext_flg = true; - if (unlikely(txq->nb_tx_avail < - ETH_TX_MIN_BDS_PER_IPV6_WITH_EXT_PKT)) - break; - } + /* TX prepare would have already checked supported tunnel Tx + * offloads. Don't rely on pkt_type marked by Rx, instead use + * tx_ol_flags to decide. + */ + tunn_flg = !!(tx_ol_flags & PKT_TX_TUNNEL_MASK); - if (RTE_ETH_IS_TUNNEL_PKT(mbuf->packet_type)) { - if (ipv6_ext_flg) { - if (unlikely(txq->nb_tx_avail < - ETH_TX_MIN_BDS_PER_TUNN_IPV6_WITH_EXT_PKT)) + if (tunn_flg) { + /* Check against max which is Tunnel IPv6 + ext */ + if (unlikely(txq->nb_tx_avail < + ETH_TX_MIN_BDS_PER_TUNN_IPV6_WITH_EXT_PKT)) break; - } - tunn_flg = true; + /* First indicate its a tunnel pkt */ bd1_bf |= ETH_TX_DATA_1ST_BD_TUNN_FLAG_MASK << ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT; @@ -1592,17 +1966,83 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) bd1_bf ^= 1 << ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT; } + /* Outer IP checksum offload */ - if (tx_ol_flags & PKT_TX_OUTER_IP_CKSUM) { + if (tx_ol_flags & (PKT_TX_OUTER_IP_CKSUM | + PKT_TX_OUTER_IPV4)) { bd1_bd_flags_bf |= ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_MASK << ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT; } - /* Outer UDP checksum offload */ - bd1_bd_flags_bf |= - ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK << - ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT; - } + + /** + * Currently, only inner checksum offload in MPLS-in-UDP + * tunnel with one MPLS label is supported. Both outer + * and inner layers lengths need to be provided in + * mbuf. + */ + if ((tx_ol_flags & PKT_TX_TUNNEL_MASK) == + PKT_TX_TUNNEL_MPLSINUDP) { + mplsoudp_flg = true; +#ifdef RTE_LIBRTE_QEDE_DEBUG_TX + qede_mpls_tunn_tx_sanity_check(mbuf, txq); +#endif + /* Outer L4 offset in two byte words */ + tunn_l4_hdr_start_offset = + (mbuf->outer_l2_len + mbuf->outer_l3_len) / 2; + /* Tunnel header size in two byte words */ + tunn_hdr_size = (mbuf->outer_l2_len + + mbuf->outer_l3_len + + MPLSINUDP_HDR_SIZE) / 2; + /* Inner L2 header size in two byte words */ + inner_l2_hdr_size = (mbuf->l2_len - + MPLSINUDP_HDR_SIZE) / 2; + /* Inner L4 header offset from the beggining + * of inner packet in two byte words + */ + inner_l4_hdr_offset = (mbuf->l2_len - + MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2; + + /* Inner L2 size and address type */ + bd2_bf1 |= (inner_l2_hdr_size & + ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK) << + ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_SHIFT; + bd2_bf1 |= (UNICAST_ADDRESS & + ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_MASK) << + ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_SHIFT; + /* Treated as IPv6+Ext */ + bd2_bf1 |= + 1 << ETH_TX_DATA_2ND_BD_TUNN_IPV6_EXT_SHIFT; + + /* Mark inner IPv6 if present */ + if (tx_ol_flags & PKT_TX_IPV6) + bd2_bf1 |= + 1 << ETH_TX_DATA_2ND_BD_TUNN_INNER_IPV6_SHIFT; + + /* Inner L4 offsets */ + if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) && + (tx_ol_flags & (PKT_TX_UDP_CKSUM | + PKT_TX_TCP_CKSUM))) { + /* Determines if BD3 is needed */ + tunn_ipv6_ext_flg = true; + if ((tx_ol_flags & PKT_TX_L4_MASK) == + PKT_TX_UDP_CKSUM) { + bd2_bf1 |= + 1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT; + } + + /* TODO other pseudo checksum modes are + * not supported + */ + bd2_bf1 |= + ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH << + ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT; + bd2_bf2 |= (inner_l4_hdr_offset & + ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK) << + ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT; + } + } /* End MPLSoUDP */ + } /* End Tunnel handling */ if (tx_ol_flags & PKT_TX_TCP_SEG) { lso_flg = true; @@ -1614,6 +2054,10 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) * and BD2 onwards for data. */ hdr_size = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len; + if (tunn_flg) + hdr_size += mbuf->outer_l2_len + + mbuf->outer_l3_len; + bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT; bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT; @@ -1634,26 +2078,44 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) } /* Descriptor based VLAN insertion */ - if (tx_ol_flags & (PKT_TX_VLAN_PKT | PKT_TX_QINQ_PKT)) { + if (tx_ol_flags & PKT_TX_VLAN_PKT) { vlan = rte_cpu_to_le_16(mbuf->vlan_tci); bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT; } /* Offload the IP checksum in the hardware */ - if (tx_ol_flags & PKT_TX_IP_CKSUM) + if (tx_ol_flags & PKT_TX_IP_CKSUM) { bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT; + /* There's no DPDK flag to request outer-L4 csum + * offload. But in the case of tunnel if inner L3 or L4 + * csum offload is requested then we need to force + * recalculation of L4 tunnel header csum also. + */ + if (tunn_flg && ((tx_ol_flags & PKT_TX_TUNNEL_MASK) != + PKT_TX_TUNNEL_GRE)) { + bd1_bd_flags_bf |= + ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK << + ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT; + } + } /* L4 checksum offload (tcp or udp) */ - if (tx_ol_flags & (PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) + if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) && + (tx_ol_flags & (PKT_TX_UDP_CKSUM | PKT_TX_TCP_CKSUM))) { bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT; - - if (ipv6_ext_flg) { - /* TBD: check pseudo csum if tx_prepare not called? */ - bd2_bf1 |= ETH_L4_PSEUDO_CSUM_ZERO_LENGTH << - ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT; + /* There's no DPDK flag to request outer-L4 csum + * offload. But in the case of tunnel if inner L3 or L4 + * csum offload is requested then we need to force + * recalculation of L4 tunnel header csum also. + */ + if (tunn_flg) { + bd1_bd_flags_bf |= + ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK << + ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT; + } } /* Fill the entry in the SW ring and the BDs in the FW ring */ @@ -1666,51 +2128,64 @@ qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts) nbds++; /* Map MBUF linear data for DMA and set in the BD1 */ - QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_dma_addr(mbuf), - mbuf->data_len); - bd1->data.bitfields = bd1_bf; + QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf), + mbuf->data_len); + bd1->data.bitfields = rte_cpu_to_le_16(bd1_bf); bd1->data.bd_flags.bitfields = bd1_bd_flags_bf; bd1->data.vlan = vlan; - if (lso_flg || ipv6_ext_flg) { + if (lso_flg || mplsoudp_flg) { bd2 = (struct eth_tx_2nd_bd *)ecore_chain_produce (&txq->tx_pbl); memset(bd2, 0, sizeof(struct eth_tx_2nd_bd)); nbds++; /* BD1 */ - QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_dma_addr(mbuf), + QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf), hdr_size); /* BD2 */ QEDE_BD_SET_ADDR_LEN(bd2, (hdr_size + - rte_mbuf_data_dma_addr(mbuf)), + rte_mbuf_data_iova(mbuf)), mbuf->data_len - hdr_size); - bd2->data.bitfields1 = bd2_bf1; - bd2->data.bitfields2 = bd2_bf2; - + bd2->data.bitfields1 = rte_cpu_to_le_16(bd2_bf1); + if (mplsoudp_flg) { + bd2->data.bitfields2 = + rte_cpu_to_le_16(bd2_bf2); + /* Outer L3 size */ + bd2->data.tunn_ip_size = + rte_cpu_to_le_16(mbuf->outer_l3_len); + } /* BD3 */ - bd3 = (struct eth_tx_3rd_bd *)ecore_chain_produce - (&txq->tx_pbl); - memset(bd3, 0, sizeof(struct eth_tx_3rd_bd)); - nbds++; - bd3->data.bitfields = bd3_bf; - bd3->data.lso_mss = mss; + if (lso_flg || (mplsoudp_flg && tunn_ipv6_ext_flg)) { + bd3 = (struct eth_tx_3rd_bd *) + ecore_chain_produce(&txq->tx_pbl); + memset(bd3, 0, sizeof(struct eth_tx_3rd_bd)); + nbds++; + bd3->data.bitfields = rte_cpu_to_le_16(bd3_bf); + if (lso_flg) + bd3->data.lso_mss = mss; + if (mplsoudp_flg) { + bd3->data.tunn_l4_hdr_start_offset_w = + tunn_l4_hdr_start_offset; + bd3->data.tunn_hdr_size_w = + tunn_hdr_size; + } + } } /* Handle fragmented MBUF */ m_seg = mbuf->next; + /* Encode scatter gather buffer descriptors if required */ - nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3); + nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3, nbds - 1); bd1->data.nbds = nbds + nb_frags; + txq->nb_tx_avail -= bd1->data.nbds; txq->sw_tx_prod++; - rte_prefetch0(txq->sw_tx_ring[TX_PROD(txq)].mbuf); bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl)); #ifdef RTE_LIBRTE_QEDE_DEBUG_TX print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags); - PMD_TX_LOG(INFO, txq, "lso=%d tunn=%d ipv6_ext=%d\n", - lso_flg, tunn_flg, ipv6_ext_flg); #endif nb_pkt_sent++; txq->xmit_pkts++; @@ -1739,3 +2214,84 @@ qede_rxtx_pkts_dummy(__rte_unused void *p_rxq, { return 0; } + + +/* this function does a fake walk through over completion queue + * to calculate number of BDs used by HW. + * At the end, it restores the state of completion queue. + */ +static uint16_t +qede_parse_fp_cqe(struct qede_rx_queue *rxq) +{ + uint16_t hw_comp_cons, sw_comp_cons, bd_count = 0; + union eth_rx_cqe *cqe, *orig_cqe = NULL; + + hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr); + sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring); + + if (hw_comp_cons == sw_comp_cons) + return 0; + + /* Get the CQE from the completion ring */ + cqe = (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring); + orig_cqe = cqe; + + while (sw_comp_cons != hw_comp_cons) { + switch (cqe->fast_path_regular.type) { + case ETH_RX_CQE_TYPE_REGULAR: + bd_count += cqe->fast_path_regular.bd_num; + break; + case ETH_RX_CQE_TYPE_TPA_END: + bd_count += cqe->fast_path_tpa_end.num_of_bds; + break; + default: + break; + } + + cqe = + (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring); + sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring); + } + + /* revert comp_ring to original state */ + ecore_chain_set_cons(&rxq->rx_comp_ring, sw_comp_cons, orig_cqe); + + return bd_count; +} + +int +qede_rx_descriptor_status(void *p_rxq, uint16_t offset) +{ + uint16_t hw_bd_cons, sw_bd_cons, sw_bd_prod; + uint16_t produced, consumed; + struct qede_rx_queue *rxq = p_rxq; + + if (offset > rxq->nb_rx_desc) + return -EINVAL; + + sw_bd_cons = ecore_chain_get_cons_idx(&rxq->rx_bd_ring); + sw_bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring); + + /* find BDs used by HW from completion queue elements */ + hw_bd_cons = sw_bd_cons + qede_parse_fp_cqe(rxq); + + if (hw_bd_cons < sw_bd_cons) + /* wraparound case */ + consumed = (0xffff - sw_bd_cons) + hw_bd_cons; + else + consumed = hw_bd_cons - sw_bd_cons; + + if (offset <= consumed) + return RTE_ETH_RX_DESC_DONE; + + if (sw_bd_prod < sw_bd_cons) + /* wraparound case */ + produced = (0xffff - sw_bd_cons) + sw_bd_prod; + else + produced = sw_bd_prod - sw_bd_cons; + + if (offset <= produced) + return RTE_ETH_RX_DESC_AVAIL; + + return RTE_ETH_RX_DESC_UNAVAIL; +}